Copilot commented on code in PR #1033:
URL: https://github.com/apache/iceberg-cpp/pull/1033#discussion_r4316399659
##########
src/iceberg/catalog/rest/rest_catalog.h:
##########
@@ -58,14 +58,19 @@ class ICEBERG_REST_EXPORT RestCatalog final
std::function<Result<std::shared_ptr<HttpClient>>(const
RestCatalogProperties&)>;
/// \brief Create a RestCatalog instance.
+ /// \param client_factory Optional HTTP client factory, called with the
initial config
+ /// and then the merged server config. When empty, uses the default HTTP
client.
+ /// \param auth_executor Optional non-owning executor for OAuth2 token
refresh. When
+ /// null, background token refresh is disabled. When non-null, it must
outlive the
+ /// catalog and all authentication sessions created by it. It must enqueue
work
+ /// asynchronously without blocking the shared timer thread.
/// \param metrics_executor Optional non-owning executor for asynchronous
REST metrics.
/// It must outlive the catalog and all tables created by it. When null,
REST metrics
/// are reported synchronously.
- /// \param client_factory Optional HTTP client factory, called with the
initial config
- /// and then the merged server config. When empty, uses the default HTTP
client.
static Result<std::shared_ptr<RestCatalog>> Make(const
RestCatalogProperties& config,
- Executor* metrics_executor
= nullptr,
- ClientFactory
client_factory = {});
+ ClientFactory
client_factory = {},
+ Executor* auth_executor =
nullptr,
+ Executor* metrics_executor
= nullptr);
Review Comment:
This change reorders `RestCatalog::Make` parameters (moving
`metrics_executor` from position 2 to 4). That’s a source-breaking API change
for existing callers that pass `metrics_executor` as the second argument.
Consider adding an overload preserving the prior signature (e.g., `Make(config,
metrics_executor, client_factory)`), delegating to the new implementation, and
(optionally) deprecating the old overload to ease migration.
##########
src/iceberg/util/scheduled_executor.cc:
##########
@@ -0,0 +1,286 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#include <condition_variable>
+#include <map>
+#include <mutex>
+#include <thread>
+#include <unordered_map>
+#include <unordered_set>
+#include <utility>
+
+#include "iceberg/util/scheduled_executor_internal.h"
+
+namespace iceberg::internal {
+
+class SharedTimer;
+class SharedTimerState;
+
+/// Per-executor registrations; all instances share the same timer thread.
+class ScheduledExecutorState
+ : public std::enable_shared_from_this<ScheduledExecutorState> {
+ public:
+ explicit ScheduledExecutorState(Executor& executor);
+ ~ScheduledExecutorState();
+
+ ScheduledExecutorState(const ScheduledExecutorState&) = delete;
+ ScheduledExecutorState& operator=(const ScheduledExecutorState&) = delete;
+
+ ScheduledTask Schedule(std::chrono::steady_clock::time_point deadline,
+ ExecutorTask task,
+ std::optional<ScheduledTaskRejection> on_rejected);
+ void Cancel(uint64_t id);
+ void Shutdown();
+
+ private:
+ friend class SharedTimerState;
+
+ void Dispatch(uint64_t id, ExecutorTask task,
+ std::optional<ScheduledTaskRejection> on_rejected);
+
+ Executor& executor_;
+ std::shared_ptr<SharedTimer> timer_;
+ std::mutex mutex_;
+ std::condition_variable cv_;
+ std::unordered_set<uint64_t> scheduled_;
+ size_t active_dispatches_ = 0;
+ bool shutdown_ = false;
+};
+
+// The timer thread retains its state until Run returns. This also allows the
last
+// timer owner to be released on the timer thread during process teardown.
+class SharedTimerState : public std::enable_shared_from_this<SharedTimerState>
{
+ public:
+ ~SharedTimerState() { Shutdown(); }
+
+ void Start() {
+ timer_thread_ = std::thread([self = shared_from_this()] { self->Run(); });
+ }
+
+ uint64_t Schedule(std::weak_ptr<ScheduledExecutorState> state,
+ std::chrono::steady_clock::time_point deadline,
ExecutorTask task,
+ std::optional<ScheduledTaskRejection> on_rejected) {
+ std::lock_guard lock(mutex_);
+ auto id = next_id_++;
+ auto [it, inserted] = tasks_.emplace(std::pair{deadline, id},
+ Task{.state = std::move(state),
+ .task = std::move(task),
+ .on_rejected =
std::move(on_rejected)});
+ by_id_.emplace(id, it);
+ cv_.notify_one();
+ return id;
+ }
Review Comment:
`SharedTimerState::Schedule` doesn’t check `shutdown_`. If `Shutdown()` has
been called, `Run()` exits and any subsequently scheduled tasks will never
dispatch (and will remain in `tasks_`/`by_id_` until destruction). A concrete
fix is to early-return a sentinel (e.g., `0`) when `shutdown_` is true (and
have callers treat that as an invalid `ScheduledTask`), or otherwise reject
scheduling explicitly after shutdown.
##########
src/iceberg/util/scheduled_executor.cc:
##########
@@ -0,0 +1,286 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#include <condition_variable>
+#include <map>
+#include <mutex>
+#include <thread>
+#include <unordered_map>
+#include <unordered_set>
+#include <utility>
+
+#include "iceberg/util/scheduled_executor_internal.h"
+
+namespace iceberg::internal {
+
+class SharedTimer;
+class SharedTimerState;
+
+/// Per-executor registrations; all instances share the same timer thread.
+class ScheduledExecutorState
+ : public std::enable_shared_from_this<ScheduledExecutorState> {
+ public:
+ explicit ScheduledExecutorState(Executor& executor);
+ ~ScheduledExecutorState();
+
+ ScheduledExecutorState(const ScheduledExecutorState&) = delete;
+ ScheduledExecutorState& operator=(const ScheduledExecutorState&) = delete;
+
+ ScheduledTask Schedule(std::chrono::steady_clock::time_point deadline,
+ ExecutorTask task,
+ std::optional<ScheduledTaskRejection> on_rejected);
+ void Cancel(uint64_t id);
+ void Shutdown();
+
+ private:
+ friend class SharedTimerState;
+
+ void Dispatch(uint64_t id, ExecutorTask task,
+ std::optional<ScheduledTaskRejection> on_rejected);
+
+ Executor& executor_;
+ std::shared_ptr<SharedTimer> timer_;
+ std::mutex mutex_;
+ std::condition_variable cv_;
+ std::unordered_set<uint64_t> scheduled_;
+ size_t active_dispatches_ = 0;
+ bool shutdown_ = false;
+};
+
+// The timer thread retains its state until Run returns. This also allows the
last
+// timer owner to be released on the timer thread during process teardown.
+class SharedTimerState : public std::enable_shared_from_this<SharedTimerState>
{
+ public:
+ ~SharedTimerState() { Shutdown(); }
+
+ void Start() {
+ timer_thread_ = std::thread([self = shared_from_this()] { self->Run(); });
+ }
+
+ uint64_t Schedule(std::weak_ptr<ScheduledExecutorState> state,
+ std::chrono::steady_clock::time_point deadline,
ExecutorTask task,
+ std::optional<ScheduledTaskRejection> on_rejected) {
+ std::lock_guard lock(mutex_);
+ auto id = next_id_++;
+ auto [it, inserted] = tasks_.emplace(std::pair{deadline, id},
+ Task{.state = std::move(state),
+ .task = std::move(task),
+ .on_rejected =
std::move(on_rejected)});
+ by_id_.emplace(id, it);
+ cv_.notify_one();
+ return id;
+ }
+
+ void Cancel(uint64_t id) {
+ std::optional<Task> cancelled;
+ {
+ std::lock_guard lock(mutex_);
+ auto it = by_id_.find(id);
+ if (it == by_id_.end()) return;
+ cancelled.emplace(std::move(it->second->second));
+ tasks_.erase(it->second);
+ by_id_.erase(it);
+ }
+ cv_.notify_one();
+ // Release callback captures outside the queue lock.
+ }
+
+ bool IsTimerThread() const {
+ return timer_thread_.get_id() == std::this_thread::get_id();
+ }
+
+ void Shutdown() {
+ Queue discarded;
+ {
+ std::lock_guard lock(mutex_);
+ if (shutdown_) return;
+ shutdown_ = true;
+ tasks_.swap(discarded);
+ by_id_.clear();
+ }
+ cv_.notify_one();
+ if (!timer_thread_.joinable()) return;
+ if (IsTimerThread()) {
+ timer_thread_.detach();
+ } else {
+ timer_thread_.join();
+ }
+ }
+
+ private:
+ struct Task {
+ std::weak_ptr<ScheduledExecutorState> state;
+ ExecutorTask task;
+ std::optional<ScheduledTaskRejection> on_rejected;
+ };
+
+ using Key = std::pair<std::chrono::steady_clock::time_point, uint64_t>;
+ using Queue = std::map<Key, Task>;
+
+ void Run() {
+ while (true) {
+ std::optional<Task> task;
+ uint64_t id;
+ {
+ std::unique_lock lock(mutex_);
+ cv_.wait(lock, [this] { return shutdown_ || !tasks_.empty(); });
+ if (shutdown_) return;
+
+ auto key = tasks_.begin()->first;
+ if (cv_.wait_until(lock, key.first, [this, key] {
+ return shutdown_ || tasks_.empty() || tasks_.begin()->first !=
key;
+ })) {
+ continue;
+ }
+ auto it = tasks_.begin();
+ id = it->first.second;
+ task.emplace(std::move(it->second));
+ by_id_.erase(id);
+ tasks_.erase(it);
+ }
+ if (auto state = task->state.lock()) {
+ state->Dispatch(id, std::move(task->task),
std::move(task->on_rejected));
+ }
+ }
+ }
+
+ std::mutex mutex_;
+ std::condition_variable cv_;
+ Queue tasks_;
+ std::unordered_map<uint64_t, Queue::iterator> by_id_;
+ uint64_t next_id_ = 1;
+ bool shutdown_ = false;
+ std::thread timer_thread_;
+};
+
+class SharedTimer {
+ public:
+ static std::shared_ptr<SharedTimer> Instance() {
+ static auto timer = std::make_shared<SharedTimer>();
+ return timer;
+ }
+
+ SharedTimer() : state_(std::make_shared<SharedTimerState>()) {
state_->Start(); }
+
+ ~SharedTimer() { state_->Shutdown(); }
+
+ uint64_t Schedule(std::weak_ptr<ScheduledExecutorState> state,
+ std::chrono::steady_clock::time_point deadline,
ExecutorTask task,
+ std::optional<ScheduledTaskRejection> on_rejected) {
+ return state_->Schedule(std::move(state), deadline, std::move(task),
+ std::move(on_rejected));
+ }
+
+ void Cancel(uint64_t id) { state_->Cancel(id); }
+
+ bool IsTimerThread() const { return state_->IsTimerThread(); }
+
+ private:
+ std::shared_ptr<SharedTimerState> state_;
+};
+
+void ScheduledTask::Cancel() {
+ if (id_.has_value()) {
+ if (auto state = state_.lock()) state->Cancel(*id_);
+ }
+ state_.reset();
+ id_.reset();
+}
+
+ScheduledExecutorState::ScheduledExecutorState(Executor& executor)
+ : executor_(executor), timer_(SharedTimer::Instance()) {}
+
+ScheduledExecutorState::~ScheduledExecutorState() { Shutdown(); }
+
+ScheduledTask ScheduledExecutorState::Schedule(
+ std::chrono::steady_clock::time_point deadline, ExecutorTask task,
+ std::optional<ScheduledTaskRejection> on_rejected) {
+ std::lock_guard lock(mutex_);
+ if (shutdown_) return {};
+ auto id = timer_->Schedule(weak_from_this(), deadline, std::move(task),
+ std::move(on_rejected));
+ scheduled_.insert(id);
+ return {weak_from_this(), id};
+}
+
+void ScheduledExecutorState::Cancel(uint64_t id) {
+ {
+ std::lock_guard lock(mutex_);
+ if (scheduled_.erase(id) == 0) return;
+ }
+ timer_->Cancel(id);
+}
+
+void ScheduledExecutorState::Shutdown() {
+ std::unordered_set<uint64_t> pending;
+ {
+ std::lock_guard lock(mutex_);
+ shutdown_ = true;
+ scheduled_.swap(pending);
+ }
+ for (auto id : pending) timer_->Cancel(id);
+ if (!timer_->IsTimerThread()) {
+ std::unique_lock lock(mutex_);
+ cv_.wait(lock, [this] { return active_dispatches_ == 0; });
+ }
+}
+
+void ScheduledExecutorState::Dispatch(uint64_t id, ExecutorTask task,
+ std::optional<ScheduledTaskRejection>
on_rejected) {
+ {
+ std::lock_guard lock(mutex_);
+ if (shutdown_ || scheduled_.erase(id) == 0) return;
+ ++active_dispatches_;
+ }
+ struct DispatchGuard {
+ ScheduledExecutorState& state;
+ ~DispatchGuard() {
+ std::lock_guard lock(state.mutex_);
+ --state.active_dispatches_;
Review Comment:
The `notify_all()` is currently called while holding `state.mutex_`. While
legal, it can cause unnecessary contention (woken waiters immediately block on
the same mutex). Prefer decrementing `active_dispatches_` under the lock,
releasing the lock, and then calling `notify_all()` afterward.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
